Before the AI Project: Five Data Questions Worth Answering

The fastest way to waste an AI budget is to point a model at data nobody trusts. Before any AI engagement, we walk clients through five questions. None of them mention AI — that's the point.

• Where does this data live, and who can actually access it? If the answer includes 'Rakesh's laptop,' start there.
• Do two systems disagree about the same fact? Pick the system of record before automating on top of either.
• Is history retained, or overwritten? Models and analytics both need the trail, not just the current state.
• What's the real quality level? Sample 50 records by hand; the error rate you find is the error rate AI will learn.
• What are we allowed to do with it? Customer data, supplier terms, and regulated records carry obligations that don't disappear because a model is involved.

The good news

Fixing these is unglamorous but fast — weeks, not years, for most SMEs. And the payoff is double: the AI initiative lands on solid ground, and ordinary reporting gets better immediately. Data groundwork is the rare project that pays off even if you never ship the model.